Federated Sensor Anomaly Detection with Autoencoder Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for anomaly detection in industrial processes using machine learning models are time-consuming, expensive, and lack accuracy due to insufficient data sharing across production sites, and require retraining at each site, compromising operational security and efficiency.
Innovation Solution
A method utilizing autoencoders and federated learning to generate local and global models, sharing anonymized model weightings and threshold values across clients, enabling accurate and secure anomaly detection without exposing sensitive data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If image data is shared across production sites via local network, then ML model training data availability improves, but operational security deteriorates due to potential external access and unwanted insight into production parameters
Solution Approach 1:
A centralized server acts as an intermediary between production sites, collecting and managing training data centrally. The server receives image data from multiple clients, trains the ML model, and distributes only the trained model back to clients, preventing direct peer-to-peer data sharing while maintaining data availability.
Solution Approach 2:
Instead of sharing original image data, the system creates and shares copies in the form of trained model parameters and weights. The server stores multiple copies of the trained model that can be distributed to any number of clients without exposing the original sensitive production data.
2Object-affected harmful factors
If ML model is retrained at each production site individually, then operational security is maintained, but training time and cost increase significantly
Solution Approach 1:
Multiple production sites are merged into a single federated training system where their data collectively trains one global model. The server aggregates training data from multiple clients to create a comprehensive model that benefits all participants, eliminating the need for separate training processes at each site.
Solution Approach 2:
The system performs preliminary centralized training at the server before deployment to clients. By pre-training the model with aggregated data from all production sites, the system eliminates the need for time-consuming local retraining at each client site, as the model is already optimized for their specific needs.
3Object-affected harmful factors
If ML model is trained with insufficient data at individual sites, then operational security is maintained, but anomaly detection accuracy deteriorates
Solution Approach 1:
The system transitions from single-site local training to multi-site federated training, adding a spatial dimension to data collection. By gathering training data from multiple production sites across different dimensions (locations, product types, manufacturing conditions), the model achieves higher accuracy while maintaining security through centralized processing.
Data Source
AI summary
A computer-implemented method for anomaly detection in sensor data includes a) generating and training a first and second local model via an autoencoder and determining a local threshold value for each local model output variable; b) transmitting local model weightings and local threshold values to a server; c) generating and training a global model via the autoencoder and determining global threshold values for global model output variables; d) transmitting the global model weightings and global threshold value to the first client, and adopting global model weightings for the first local model; e) capturing first sensor data by a first sensor; f) applying the first sensor data to the first local model and determining the first local model output variable of the first client; and g) detecting an anomaly for the sensor data, if the local model output variable is outside a range that is fixed by the global threshold value.